Hybrid Transformer for Anomaly Detection on Railway HVAC Systems Through Feature Ensemble of Spatial–Temporal with Multi-channel GADF Images

The Heating, Ventilating, and Air Conditioning (HVAC) system, responsible for maintaining a comfortable indoor environment in buildings and vehicles, is designed to regulate factors such as temperature, humidity, and airflow. In transportation facilities, where HVAC systems are installed, they control and adjust the temperature, humidity, and intake of air within interior spaces, ensuring a pleasant environment, enhancing service quality, and safeguarding user health. However, these systems, which are both large-scale and highly complex, pose challenges when it comes to timely and effective problem-solving using a run-to-fail policy-based reactive maintenance approach. Therefore, there is a growing demand for deep learning models capable of HVAC fault detection to design preventive management using AI. These deep learning models typically receive input in the form of time-series data generated by various sensors in HVAC systems or images generated through transformation algorithms that intuitively represent time-series data. However, conventional deep learning models for analyzing these data types tend to focus on either temporal or spatial features. Hence, to utilize both temporal and spatial features simultaneously, we propose a transformer-based hybrid model. The proposed model leverages the Gramian Angular Difference Field algorithm, one of the image transformation algorithms, to convert multivariate time-series data from HVAC into multi-channel images. It then performs each second anomaly detection by combining feature information extracted through encoders from both transformer and vision transformer. In experimental datasets, the anomaly detection performance achieved an F1 score of 0.9965. As a result, this demonstrates that the proposed deep learning model can enhance the overall reliability and safety of HVAC systems, and showing that deep learning models can serve as crucial toolsin enhancing the maintenance and safety of HVAC systems.